# AI··Rooms: Turning prompts into powerful actions

import { Steps } from "@astrojs/starlight/components"

An **AI··Room** is an auditable chat surface where you ask for information and actions in natural language. You mention
services (credentials in the environment) or flows with `@`, review what the agent proposes, and approve or deny
actions that change systems. Studio’s sidebar label is **Rooms**. Screenshots on this page use IP reputation and ticket
follow-ups so you can see mentions, replies, and approval prompts.

## When to use

- You need ad-hoc investigation or actions across tools without building a full flow first.
- Analysts should stay in one conversation while checking reputation, tickets, or cloud identity data.
- Prefer [Run a flow from an AI··Room](../../build/trigger-your-flow/run-a-flow-from-an-ai-room/) when a repeatable
  **Chathook** flow should start from `@flowName` and optionally reply in the room.
- Prefer designer flows with schedules or webhooks when the job is predictable and should run without chat.

Before you begin, pick the **environment** that holds the credentials (and deployed flows) you will mention. You need
access to **Rooms** in that environment.

## Guide

### Open a room

Rooms live under **Build & operate** in the sidebar. Each conversation is tied to an environment so mentions resolve to
that environment’s credentials and flows.

<ol>
<li>

**Open Rooms.** In the left sidebar, select **Rooms**.

</li>
<li>

**Choose the environment.** In the rooms sidebar, select the environment whose credentials and flows you will use.

</li>
<li>

**Start or rename the room.** A new conversation starts as **New room**; rename it when you want a clear label for the
thread.
<div class="docs-screenshot-compact">

![Rooms page with New room header, Chat with your tools prompts, and Ask your tool anything input](./assets/rooms-page-new-room-header-chat.png)

</div>

</li>
</ol>
### Mention services and ask in natural language

Use `@` to mention a connected service, then describe what you want. Studio searches service titles and descriptions.
You can also mention flows that are callable from rooms (for example with a Chathook).

<ol>
<li>

**Type `@` and pick a service.** Choose the integration, then complete the request in plain language.
<div class="docs-screenshot-compact">

![Room composer showing an @Atlassian Jira mention and a request for the most recent ticket in a project](./assets/room-composer-showing-atlassian-jira-mention.png)

</div>

</li>
<li>

**Review the reply.** The agent returns the operations it used and the response body so you can verify the result.
<div class="docs-screenshot-compact">

![Room reply showing operations performed and the response body for a Jira ticket lookup](./assets/room-reply-showing-operations-performed-response.png)

</div>

</li>
<li>

**Use suggested follow-ups.** When the agent offers a button (for example to add a comment), it can fill a follow-up
prompt for you to send.
<div class="docs-screenshot-compact">

![Room suggestion button to add a comment on a Jira issue](./assets/room-suggestion-button-add-comment-jira.png)

</div>
<div class="docs-screenshot-compact">

![Pre-filled follow-up prompt in the room composer after choosing a suggested action](./assets/pre-filled-follow-up-prompt-room.png)

</div>

</li>
</ol>
### Approve or deny actions

When the agent needs to change something in a connected system, it asks for your approval before it continues. You
**Approve** or **Deny** (or approve/deny all when several actions are pending).

<ol>
<li>

**Review the proposed action.** Read what will run, then choose **Approve** or **Deny**.
<div class="docs-screenshot-compact">

![Room approval prompt asking the operator to approve or deny a proposed agent action](./assets/room-approval-prompt-asking-operator-approve.png)

</div>

</li>
</ol>
## Example

An analyst opens **Rooms** in the **Prod** environment and starts a room to triage a suspicious URL. They mention
`@VirusTotal`, check reputation, then follow a suggested prompt to post a short summary to Slack. When the agent
proposes a write action, they approve it only after checking the target channel. For a repeatable IP look-up that should
always run the same flow from chat, they use a Chathook instead (see Related).

## Keep in mind

- Mentions resolve against the **selected environment**—credentials and flows must exist there.
- Treat agent output as assistance; verify important results in the source system when the stakes are high.
- Do not paste secrets or personal data into the room that you would not store in other Studio fields.

## Related

- [Run a flow from an AI··Room](../../build/trigger-your-flow/run-a-flow-from-an-ai-room/)
- [Operate](../)